Pith. sign in

REVIEW 5 major objections 7 minor 1 cited by

Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?

T0 review · 5 major / 7 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper argues that RL-trained reasoning models can match test-set-trained 'oracle' models while trained only on the training split, so held-out accuracy no longer measures RL generalization, and stress tests show the learned skills are

desk verdict The OPG measurement is a genuinely new and useful diagnostic, but the strong conclusion that benchmarks are broken rests on an asserted criterion and an oracle that may be undertrained. read the letter →

arxiv 2510.10541 v2 pith:QHH4A43R submitted 2025-10-12 cs.LG cs.AI

Zihan Chen , Yiming Zhang , Hengguang Zhou , Zenghui Ding , Yining Sun , Cho-Jui Hsieh This is my paper · ORCID
classification cs.LGcs.AI
keywords reinforcementlearningLLMreasoningbenchmarkevaluationgeneralizationgapOraclePerformancedistributionshiftcounterfactualdifficultystratification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that current benchmarks for evaluating RL-trained LLM reasoning are structurally incapable of revealing how well these models generalize. The key evidence is that an RL model trained on a benchmark's training split scores about the same as an RL model trained directly on the test split, so the test set's 'unseen-ness' no longer acts as a meaningful obstacle. The paper then shows that high benchmark scores coexist with brittle skills: specialists trained on easy problems fail on hard ones, narrow-distribution fine-tuning can make the model worse than baseline on far-out-of-distribution data, and switching a problem to a counterfactual rule causes a large accuracy drop. It concludes with three principles for faithful benchmarks: difficulty stratification, distributional robustness, and counterfactual reasoning.

What carries the argument

The central object is the Oracle Performance Gap (OPG), the normalized difference in pass@1 accuracy between an 'oracle' model fine-tuned directly on the benchmark test set and a standard model fine-tuned only on its train set. A vanishing OPG is taken as the diagnostic signature of a benchmark that no longer tests generalization. Supporting tools are the difficulty-stratified cross-generalization analysis, a semantically clustered distribution test, and a counterfactual test that rewrites a core rule to force a choice between recitation and on-the-fly reasoning.

What would settle it

Run the same train-versus-test OPG comparison on a test set that is semantically disjoint from training (for example, generated with a held-out problem-generation routine). A large positive OPG on that version would show the vanishing gap is a property of the current benchmark's homogeneity rather than of RL in general.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the conventional train/test split no longer separates RL-trained reasoning models: the Oracle Performance Gap—the normalized accuracy difference between a model fine-tuned on the test split and one fine-tuned on the training split—is essentially zero (roughly -5% to +1.5% across the benchmarks), while SFT shows a substantial gap (up to 36%). The paper interprets this as 'unseen-ness' no longer being a diagnostic signal for RL generalization, and reinforces it with three stress tests: performance collapses for easy-trained specialists on harder problems; semantically narrow fine-tuning can invert to a penalty on far out-of-distribution data; and counterfa

Load-bearing premise

The load-bearing premise is that a useful benchmark must show a clearly positive OPG; if a benchmark could separate strong from weak RL policies while train-trained and test-trained models scored alike, the paper's central conclusion would not follow.

Editorial extensions

If this is right

  • Benchmark scores reported with a single train/test split can no longer be read as evidence of RL generalization; a model can match near-oracle scores without having seen the test questions.
  • Future benchmarks should report difficulty-stratified results, because aggregate accuracy masks large asymmetries in which specialist models trained on hard problems transfer while easy-trained models do not.
  • Benchmarks should include distribution-shift sets; the observed performance inversion shows fine-tuning can actively harm performance on sufficiently distant data.
  • Benchmarks should include counterfactual rewrites; the accuracy drop shows models default to recitation rather than following novel premises.
  • Training data difficulty matters: models trained on harder problems have higher cross-difficulty transfer, so benchmark training sets should be sufficiently challenging.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • My inference: a low OPG could also be produced by the base model having encountered near-duplicate problems during pretraining, so a near-zero gap alone may indicate contamination rather than a genuine absence of generalization.
  • My inference: the difficulty-stratified result suggests a concrete training recipe—curricula should front-load hard problems—though the paper itself stops short of advocating this.
  • My inference: the three principles could be turned into a composite benchmark-validity metric that combines OPG, difficulty-decay slope, OOD inversion margin, and counterfactual drop, allowing the community to compare benchmarks rather than only model scores.
  • My inference: a direct follow-up would test whether process-based rewards or explicit rule-following training shrink the counterfactual drop; the paper does not experiment with those alternatives.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 7 minor

Summary. The paper proposes the Oracle Performance Gap (OPG), defined in Eq. (1) as the normalized difference between a model trained on the test split and a model trained on the train split. Across MATH, GSM8K, HeadQA, and DeepScaler with Qwen2.5-3B/7B models, the authors report that OPG is near zero for GRPO-trained RL models but large for SFT models, concluding that unseen-ness is no longer a valid measure of RL generalization. The paper then introduces three stress tests (difficulty stratification, distributional shift, and counterfactual reasoning) to show that RL-learned skills are brittle, and it proposes three benchmark design principles: difficulty stratification, distributional robustness, and counterfactual reasoning.

Significance. If the central claim were well supported, the paper would make an important contribution to RL evaluation for LLMs: it offers a concrete metric (OPG), a multi-benchmark study, and a set of actionable design principles. The difficulty-stratified analysis is a useful idea, and the distribution/counterfactual tests are creative diagnostics. However, the empirical support is currently insufficient. The OPG interpretation depends on an unverified oracle, the headline tables report single runs without variance, the distribution test has severe sampling noise, the counterfactual test lacks a baseline, and the central interpretive standard (Eq. 2) is asserted rather than validated. These issues are load-bearing for the paper's main conclusion, so the paper cannot be accepted in its current form. With additional experiments and careful statistical treatment, the core ideas could become a valuable contribution.

major comments (5)
  1. [§2.1.2 / Appendix A.1] The RL oracle M_RL,test is defined as a GRPO model trained on the test split with the same setup, but no convergence evidence is provided. With only 5 epochs and a KL coefficient of 1e-2, the oracle may be under-trained or over-regularized, unable to exploit the test set even in principle. The paper never reports the oracle's training-set accuracy or learning curves. This is critical because OPG≈0 is interpreted as 'unseen-ness is insufficient'; if the oracle is not a genuine upper bound, the OPG is artificially small. Please report learning curves, training-set accuracy, and verify that longer training or a lower KL coefficient does not substantially improve test accuracy.
  2. [§2.2 / Tables 1–2] The headline OPG tables report single runs without variance. Many OPG values are sub-1pp (e.g., 0.31, -0.54, 0.73, 0.89) and are within binomial sampling error for the reported accuracies. More importantly, Table 1 contains a DeepScaler 7B OPG of -5.07%, meaning the train-trained model beats the oracle by 5% relative; this directly contradicts the claim that OPG is 'consistently low' across benchmarks. Please report multiple seeds, confidence intervals, and explicitly discuss the DeepScaler 7B outlier.
  3. [§3.2.1 / Table 3] The distribution test uses only 80 samples per distance bin. The performance gains at d2–d5 are +0.75, 0.00, -1.25, -2.50 percentage points. For n=80 and accuracies around 45%, the standard error of the difference is roughly 7–8 percentage points, so none of these values is statistically distinguishable from zero. The claimed 'performance inversion' in Finding 4 is not supported by the data as reported. Please provide confidence intervals, increase sample size, or run multiple seeds and report the distribution of gains.
  4. [§3.2.2 / Table 4] The counterfactual test lacks a baseline. The paper reports only RL-trained models on D_bal and D_cf. Without measuring the base model or an SFT model on the same D_cf, one cannot attribute the drop from 64.2/74.8 to 36.0/41.2 to RL training, to the counterfactual transformation, or to a general inability to follow the novel premise. Please add such baselines and also report the size of D_cf.
  5. [§2.1.1 / Eqs. (2)–(3)] The central interpretive claim is that OPG≫0 defines a benchmark that effectively measures generalization. This threshold is asserted, not derived or empirically validated. A benchmark could have OPG≈0 and still be highly discriminative among RL policies if training on the train split already captures the test distribution; conversely, a large OPG could reflect oracle overfitting rather than a good benchmark. To make the paper's conclusion stand, the authors should demonstrate that OPG correlates with an independent measure of benchmark quality, such as the ability to rank known strong versus weak policies. Without this, the paper's main conclusion is an unvalidated interpretive standard applied to its own measurements.
minor comments (7)
  1. [§2 title] 'Generalizatgion' is a typo in the section heading.
  2. [Appendix E.2] The worked counterfactual example contains contradictory numbers: the correct answer is given as both 3 and 9, and the model's answer as both 1 and 14/3. These inconsistencies undermine the qualitative evidence.
  3. [Figure 2] The axis labels contain 'uni00A0' artifacts; the Unicode issues should be fixed.
  4. [Appendix D] Core training set selection uses t-SNE coordinates, which are stochastic. The t-SNE random seed and exact parameters must be reported for reproducibility; otherwise the protocol is not reproducible.
  5. [Appendix B.1] The difficulty annotation is performed by a single LLM (Gemini 2.5 Pro) with no validation. Report inter-annotator agreement or a human-reviewed sample to support the rubric's reliability.
  6. [§3.1.1 / Figure 2(a)] The 'oracle gap' in Figure 2(a) appears to be a different quantity from the OPG in Eq. (1). Define it explicitly to avoid confusion.
  7. [Table 4] The sample sizes for D_bal and D_cf are not reported; please include them.

Circularity Check

1 steps flagged · score 2.0 of 10

OPG's benchmark-failure criterion is definitional; stress tests are independent, so circularity is minor.

  1. self definitional [Section 2.1.1, Eq. (3); Section 2.2, Takeaway]
    "2.A Benchmark that Fails to Measure Generalization yields nearly identical performance between models trained on the training split and those trained on the test split. This is reflected by a negligible OPG: OPGA ≲ 0."

    The paper defines 'benchmark fails to measure generalization' as OPG ≲ 0 (Eq. 3), then measures OPG ≈ 0 and concludes that 'test-set unseen-ness is no longer a valid measure of generalization.' The conclusion is the definition restated: no independent standard establishes that small OPG implies benchmark failure. The empirical measurement of OPG is real and independent, but its interpretation as a benchmark-failure signal is imposed by Eq. (2)-(3), not derived from the data.

full rationale

The central OPG result is partly definitional: the paper equates a negligible OPG with benchmark failure, so the headline conclusion that 'unseen-ness alone is no longer the challenging or discriminative criterion' is the metric definition applied to a measurement rather than an independent derivation. The measurements themselves (Tables 1 and 2) are empirical and reproducible, and the stress tests (difficulty, distribution, counterfactual) are independent of the OPG definition and do not reduce to fitted inputs. The self-citation to the authors' own FRAME work appears only in a list of reasoning strategies and is not load-bearing. The appendix's RL oracle uses 5 epochs and KL regularization, which raises an empirical validity concern about whether M_RL,test is a true upper bound, but that is a correctness issue, not circularity. Overall, only the definitional interpretive step prevents a score of 0; there is no fitted-input-called-prediction or self-citation-chain circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on an asserted interpretation of OPG, on LLM-generated counterfactual transformations, and on a t-SNE-based notion of semantic distance; none of these is independently benchmarked, but the underlying fine-tuning and evaluation are standard, so these are domain assumptions rather than invented entities.

free parameters (3)
  • k (number of semantic clusters) = 3
    Chosen via Elbow/Silhouette; determines the semantic clusters from which core and OOD test sets are built (Appendix D).
  • Number of distance bins and samples per bin = 5 bins; 80 samples each
    Hand-chosen; n=80 per bin makes the distribution-test inversion statistically fragile (Table 3).
  • Difficulty annotation rubric = 5 levels L1-L5
    Hand-defined cognitive-complexity rubric applied via Gemini 2.5 Pro; all difficulty-stratified conclusions depend on it (Appendix B.1).
assumptions (4)
  • domain assumption A benchmark that effectively measures generalization must present a substantial challenge reflected by OPG ≫ 0 (Eq. 2); equivalently, a near-zero OPG means the benchmark fails.
    This is the load-bearing interpretation of the OPG metric. It is asserted (Section 2.1.1) rather than derived.
  • domain assumption The Gemini-generated counterfactual transformations only alter the target rule and preserve difficulty, so any drop on D_cf reflects recitation rather than artifact difficulty.
    Section 3.2.2 and Appendix E rely on this to interpret performance collapse as 'recitation over reasoning'.
  • domain assumption Euclidean distances in a 2D t-SNE projection approximate semantic distance for binning OOD test sets.
    Appendix D uses 2D t-SNE coordinates for nearest-neighbor selection and distance bins although t-SNE distances are not generally faithful to high-dimensional geometry.
  • domain assumption Single-run pass@1 differences are meaningful beyond sampling noise; no error bars or repeated seeds are reported.
    Tables 1-4 report single pass@1 values and treat differences below 1pp as structural signals.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?." pith.science (2026). https://pith.science/paper/QHH4A43R

@misc{pith2026251010541,
  author       = {Pith},
  title        = {Pith review of: Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QHH4A43R}},
  note         = {Machine review of arXiv:2510.10541}
}
read the original abstract

Current benchmarks are inadequate for evaluating progress in reinforcement learning (RL) for large language models (LLMs).Despite recent benchmark gains reported for RL, we find that training on these benchmarks' training sets achieves nearly the same performance as training directly on the test sets, suggesting that the benchmarks cannot reliably separate further progress.To study this phenomenon, we introduce a diagnostic suite and the Oracle Performance Gap (OPG) metric that quantifies the performance difference between training on the train split versus the test split of a benchmark. We further analyze this phenomenon with stress tests and find that, despite strong benchmark scores, existing RL methods struggle to generalize across distribution shifts, varying levels of difficulty, and counterfactual scenarios: shortcomings that current benchmarks fail to reveal.We conclude that current benchmarks are insufficient for evaluating generalization and propose three core principles for designing more faithful benchmarks: sufficient difficulty, balanced evaluation, and distributional robustness.

Figures

Figures reproduced from arXiv: 2510.10541 by the authors.

Figure 1
Figure 1. Overview of our empirical framework. The workflow begins by diagnosing benchmark flaws with novel metrics to uncover a core symptom: a vanishing generalization gap. It then proceeds through a suite of stress tests that reveal the brittle, shortcut-based nature of the learned skills, culminating in a new set of principles for more robust evaluation. gap observed in RL agents. Building on this, we subject RL-tuned mod… view at source ↗
Figure 2
Figure 2. The Illusion of Average Performance. (a) The mean performance gap between the best (specialist) model and the average of all other models widens dramatically as task difficulty in￾creases. (b) Surprisingly, the average scores of these specialists (calculated across all five diffi￾culty partitions) are nearly identical. This contrast illustrates how a difficulty-agnostic evaluation can mask substantial differences in… view at source ↗
Figure 3
Figure 3. The Average Cross-Difficulty Generalization score for 3B and 7B models. The y-axis represents the average accuracy of a specialist model (trained on level Li) on all other, unseen difficulty levels. Both models show a clear trend: as the training data complexity in￾creases, the model’s ability to generalize to other diffi￾culties improves, with the Level 5-trained model being the strongest generalist. Finding 3: Tra… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: in the main text are directly derived from the raw accuracy scores presented here [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the experimental data construction for the distribution test. (a) The highly concentrated Dcore set is selected from a semantic cluster. (b) The test sets are sampled and binned based on their increasing semantic distance from the Dcore centroid. E THE…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    Argues for shifting to diagnosis-driven tension management of offline priors in online RL, supported by a framework on prior roles, experiments showing help-or-hurt reversals, and cross-domain evidence.

Reference graph

Works this paper leans on

25 extracted references · 10 linked inside Pith · cited by 1 Pith paper

  1. [1]

    Training verifiers to solve math word problems

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021

    arXiv 2021

  2. [2]

    Shortcut learning in deep neural networks

    Robert Geirhos, J \"o rn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2 0 (11): 0 665--673, 2020

    2020

  3. [3]

    Measuring massive multitask language understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. Proceedings of the International Conference on Learning Representations (ICLR), 2021

    2021

  4. [4]

    Adversarial examples for evaluating reading comprehension systems

    Robin Jia and Percy Liang. Adversarial examples for evaluating reading comprehension systems. arXiv preprint arXiv:1707.07328, 2017

    arXiv 2017

  5. [5]

    Solving quantitative reasoning problems with language models, 2022

    Aitor Lewkowycz, Anders Andreassen, David Dohan, Ethan Dyer, Henryk Michalewski, Vinay Ramasesh, Ambrose Slone, Cem Anil, Imanol Schlag, Theo Gutman-Solo, et al. Solving quantitative reasoning problems with language models, 2022. URL https://arxiv. org/abs/2206.14858, 1, 2022

    arXiv 2022

  6. [6]

    Think or not think: A study of explicit thinking in rule-based visual reinforcement fine-tuning

    Ming Li, Jike Zhong, Shitian Zhao, Yuxiang Lai, Haoquan Zhang, Wang Bill Zhu, and Kaipeng Zhang. Think or not think: A study of explicit thinking in rule-based visual reinforcement fine-tuning. arXiv preprint arXiv:2503.16188, 2025

    arXiv 2025

  7. [7]

    Let's verify step by step

    Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. Let's verify step by step. In The Twelfth International Conference on Learning Representations, 2023

    2023

  8. [8]

    Deepscaler: Surpassing o1-preview with a 1.5 b model by scaling rl, 2025

    Michael Luo, Sijun Tan, Justin Wong, Xiaoxiang Shi, William Y Tang, Manan Roongta, Colin Cai, Jeffrey Luo, Tianjun Zhang, Li Erran Li, et al. Deepscaler: Surpassing o1-preview with a 1.5 b model by scaling rl, 2025. Notion Blog, 3 0 (4): 0 5

    2025

Show all 25 references
  1. [9]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35: 0 2...

    2022

  2. [10]

    Curriculum reinforcement learning from easy to hard tasks improves llm reasoning

    Shubham Parashar, Shurui Gui, Xiner Li, Hongyi Ling, Sushil Vemuri, Blake Olson, Eric Li, Yu Zhang, James Caverlee, Dileep Kalathil, et al. Curriculum reinforcement learning from easy to hard tasks improves llm reasoning. arXiv preprint arXiv:2506.06632, 2025

    2025

  3. [11]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in neural information processing systems, 36: 0 53728--53741, 2023

    2023

  4. [12]

    Proximal policy optimization algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017

    2017 arXiv

  5. [13]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300, 2024

    2024 arXiv

  6. [14]

    Head-qa: A healthcare dataset for complex reasoning

    David Vilares and Carlos G \'o mez-Rodr \' guez. Head-qa: A healthcare dataset for complex reasoning. arXiv preprint arXiv:1906.04701, 2019

    1906 arXiv

  7. [15]

    Self-consistency improves chain of thought reasoning in language models

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171, 2022

    2022 arXiv

  8. [16]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 0 24824--24837, 2022

    2022

  9. [17]

    Tree of thoughts: Deliberate problem solving with large language models, 2023

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L Griffiths, Yuan Cao, and Karthik Narasimhan. Tree of thoughts: Deliberate problem solving with large language models, 2023. URL https://arxiv. org/abs/2305.10601, 3: 0 1, 2023

    2023 arXiv

  10. [18]

    Frame: Feedback-refined agent methodology for enhancing medical research insights

    Chengzhang Yu, Yiming Zhang, Zhixin Liu, Zenghui Ding, Yining Sun, and Zhanpeng Jin. Frame: Feedback-refined agent methodology for enhancing medical research insights. In Annual Meeting of the Association for Computational Linguistics, 2025. URL https://api.semanticscholar.org...

    2025

  11. [19]

    Simplerl-zoo: Investigating and taming zero reinforcement learning for open base models in the wild

    Weihao Zeng, Yuzhen Huang, Qian Liu, Wei Liu, Keqing He, Zejun Ma, and Junxian He. Simplerl-zoo: Investigating and taming zero reinforcement learning for open base models in the wild. arXiv preprint arXiv:2503.18892, 2025

    2025 arXiv

  12. [20]

    Llamafactory: Unified efficient fine-tuning of 100+ language models

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. Llamafactory: Unified efficient fine-tuning of 100+ language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System ...

    2024 arXiv

  13. [21]

    Easyr1: An efficient, scalable, multi-modality rl training framework

    Yaowei Zheng, Junting Lu, Shenzhi Wang, Zhangchi Feng, Dongdong Kuang, and Yuwen Xiong. Easyr1: An efficient, scalable, multi-modality rl training framework. https://github.com/hiyouga/EasyR1, 2025

    2025

  14. [22]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  15. [23]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  16. [24]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  17. [25]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.